Bank Runs Without Sunspots

Francisco J. Santos-Arteaga
WP02/10

Resumen
La literatura sobre pnicos bancarios reduce todo mecanismo de coordinacin que de lugar a los ataques a puntos solares exgenos. Presentamos una versin de equilibrio general de dichos modelos en la cual la incertidumbre a la que los agentes econmicos se hallan sujetos es modelizada de forma explcita, de manera que los pnicos surgen como resultados ptimos de equilibrio correspondientes a juegos de coordinacin Bayesianos jugados por agentes racionales antes de depositar sus fondos en el banco. Diferencias en los conjuntos de informacin entre el banco y sus clientes dan lugar a pnicos racionales autocontenidos en equilibrio. La coexistencia de diversas creencias probabilsticas en equilibrio, as como la naturaleza auto-contenida de los ataques, derivan del principio de la mano invisible de Adam Smith. Los pnicos obtenidos no violan el principio de revelacin.

Palabras clave: pnicos bancarios, ataques autocontenidos, juegos de coordinacin Bayesianos, principio de revelacin, principios de la mano invisible.

Abstract
The literature on bank runs reduces all coordination mechanisms triggering attacks on banks to exogenous sunspots. We present a general equilibrium version of these models where the uncertainty faced by depositors is modeled explicitly, such that bank runs arise as optimal equilibrium outcomes corresponding to Bayesian coordination games played by rational agents before depositing. Differentials in information sets between the bank and its depositors lead to rational self-contained equilibrium runs. The coexistence of different beliefs in equilibrium jointly with the self-fulfilling nature of the attacks follow from Adam Smith's invisible hand principle. The runs obtained do not violate the revelation principle.

Key words: bank runs, self-contained attacks, Bayesian coordination games, revelation
principle, invisible hand principle.

The author thanks Debora Di Caprio, and seminar participants at the Universidad de Salamanca and Universidad del Pais Vasco for helpful discussions and comments. Tel. +(39)0471.013.288 Fax. +(39)0471.013.009 e-mail: FSantosArteaga@unibz.it Universidad Complutense de Madrid Campus de Somosaguas, 28223-Pozuelo Instituto Complutense de Estudios Internacionales, Universidad Complutense de Madrid. Campus de Somosaguas, Finca Mas Ferre. 28223, Pozuelo de Alarcn, Madrid, Spain.
Francisco J. Santos-Arteaga. ISBN: 978-84-692-9959-3 Depsito legal:
El ICEI no comparte necesariamente las opiniones expresadas en este trabajo, que son de exclusiva responsabilidad de sus autores

Index
1. 2. Motivation and Stylized Facts ...7 Theoretical Strategic Behavior Locational advantages..........9 2.1. 2.2. 2.3. 3. Initial assumptions..........................................................................................9 Contract-induced coordination games..............................................................10 Updating the bayesian beliefs of depositors....11

Post-Deposit Coordination Games ..13 3.1. 3.2. 3.3. 3.4. 3.5. Pre and post-deposit equilibria....13 The set of post-deposit games......13 Communication strategies...15 Post-deposit equilibria set....16 Mechanism monotonicity and self-contained run equilibria.18

4.

Main Results..21 4.1. Numerical proofs..22

5.

A note on Suspension of Convertibility...25 Annex 1.. ...27

.

Bibliographical references.28

1. Motivation and Stylized Facts

There exists a recognized relationship between currency crises and bank runs, the latter preceding the former, defining the most recent financial speculative episodes in economic history, namely the Mexican and Asian crises. Kaminsky and Reinhart (1999) provide ample empirical evidence illustrating this causality. The second generation currency crises literature, see Cole and Kehoe (2000), developed after Obstfeld's (1994) seminal paper, emphasizes the behavior of agents as the driving force behind the attacks in their models. Though multiple equilibria are allowed, and the one leading to an attack is generated by an exogenous sunspot variable, rationality prevails on the expectations formation process of agents and safe areas based on the value of fundamentals are defined. The literature on bank runs, on the other hand, relies heavily on 1 sunspot variables to explain the phenomenon, or eliminates it from the equilibrium out2 comes. The strategic behavior of agents is generally recognized but not studied explic3 itly. In order to explain the above empirical relationship, bank runs and currency crises were both initially modeled as sunspot phenomena. Since it seems plausible to assume that most of the agents participating in the currency attacks were also involved in the preceding bank runs, both types of models provided a consistent explanation for the entire cycle defining speculative attacks. An exogenous sunspot affects a set of depositors, who run on their corresponding banks, destabilizing the financial structure of the economy. This is followed by an attack on the currency, by mostly the same agents who run previously on the bank, which further destabilizes the banking system, causing additional runs, and so on. Even though fundamentals play a crucial role modifying the expectations of agents through the different currency attack areas, the attack is ultimately
See, for example, Diamond and Dybvig (1983), and Peck and Shell (2003). 2 Green and Lin (2000) and (2003), design sequentially efficient run-proof contracts, but their model is subject to a monitoring the monitor problem in the sense defined by Krasa and Villamil (1992). 3 Exceptions to this rule are Chari and Jagannathan (1988) and Samartin (2003). However, the set of assumptions needed to generate their results, consisting of confounding multiple equilibria, may seem too restrictive and applicable only in particular cases.
1

generated by a sunspot variable that coordinates agents on the corresponding equilibrium. This idea gained acceptance due to the fact that recent speculative episodes took place when fundamentals were not bad enough to justify rational attacks on the currency, while no attacks had occurred when fundamentals displayed relatively worse values. The sunspot hypothesis defining sequential self-reinforcing attacks on the banks and currency of a country provided a coherent explanation for both phenomena until Morris and Shin published their paper in 1998. Speculative attacks on currencies were no longer defined uniquely by sunspot variables, but were allowed to be the consequence of rational behavior coupled with imperfect information and 4 noisy signals. This result opened a theoretical paradox that has not been closed yet. Assume that agents are rational utility maximizers who have access to a given information set that is updated at each point in time. These agents, in their role as depositors, run on the bank if, and only if, the attack is coordinated on the realization of an exogenous sunspot variable, for a given value of the fundamentals. Bank runs are followed by speculative attacks on the currency of the corresponding country. However, as shown by Morris and Shin, attacks on the currency are based on signals about the state of the economy received by depositors after running on the bank. The financial system will send a negative self-fulfilling signal, which derives directly from the run triggered by the same agents observing the signal, on which the decision of whether or not to attack the currency will be based. The positive cyclical behavior described above implies that theoretically rational agents stop the attack on the currency to analyze a set of signals that they have just generated by previously running on the bank, based on an exogenous sunspot variable. That is, the current state of the literature defines rational attacks on a currency that cannot be generated by sunspot-based runs on the banks of the corresponding country. The main purpose of the current paper is to solve this paradox by defining runs on the financial intermediaries of a country as rational equilibria independent of any sunspot variable realiWe are obviously referring to the attacks that cannot be justified by the value of fundamentals.
4

zation. We consider explicitly the information structure defined within the standard bank run model, and allow for beliefs differentials to exist between the bank and its depositors. Even though the strategic framework used by the second generation currency crises literature is almost identical to the one employed to model bank runs, two important differences must be considered. First, excluding sunspot realizations, the information set of the agents modelled in the bank run literature is fixed 5 through time, while currency speculators are allowed to observe signals before making a decision. Second, the coordination game that speculators play before attacking a currency is ignored by depositors when deciding whether to run on the bank or not. The second difference deserves some additional comments. Agents do play a coordination game when deciding whether to deposit their funds in the bank or to remain in an autarky situation. This game has a unique Nash equilibrium in pure strategies where agents deposit their funds and decide ex-ante not to 6 run on the bank. The existence of a financial intermediary signals that a coordinated no-run equilibrium has been reached among depositors in the deposit game. Thus, it should be clear that the same pre-deposit coordination game cannot be used to model the post-deposit run coordination problem to which agents are subject after depositing funds. While this is valid only if the information set of depositors remains unchanged, it is incorrect if agents are allowed to update their information sets by monitoring the bank. In other words, if the equilibrium of the game has been chosen by the agents beforehand, there is no reason to expect a change unless motivated by an exogenous coordination variable. This point has already been made by Alonso (1996) and Jacklin and Bhattacharya (1988), where postThis is not the case in Alonso (1996) and Jacklin and Bhattacharya (1988), where depositors receive signals on which to base their withdrawing decision. In this case, runs are entirely defined by the state of fundamentals. We explain below why we follow a different approach to model bank runs. 6 The game has an additional equilibrium in mixed strategies as shown by Adao and Temzelides (1998), but it does not prevail if a strong form of forward induction requirement is imposed on the equilibrium. Postlewaite and Vives (1987) showed using a similar game that equilibria involving a positive run probability exist and are unique. They define three different types of agents, as opposed to the general case with only two types, depending on their degree of impatience. The most patient type does not misrepresent her type even though she knows that less patient types will with probability one. This is the case since their payoff, though affected by early withdrawals, is still higher under truth telling.
5

deposit signals defining withdrawal decisions are received by depositors. These models assume that signals reflect the true state of fundamentals, leading to runs caused entirely by negative shocks to the economy. That is, bank runs and the subsequent currency attacks would be mainly due to bad fundamentals, as the first generation currency crises literature predicts, see Krugman (1979). However, this result is at odds with the empirical irrationality exhibited by agents in the crises episodes under consideration, where attacks were coupled with relatively good values of the fundamental 7 variables. We present a general equilibrium model where information spreads regarding the expected state of the economy generate rational selfcontained bank runs. Information differentials between the bank and its depositors lead to the design of incentive compatible, statecontingent, and sequential demand deposit contracts that are subject to self-contained runs by depositors, whose information sets are defined by mean preserving spreads on the true expected state observed by the bank. These contracts dominate the autarky allocation, meaning that agents deposit funds even if they expect an attack to take place in (at least) one of the states of the world. Depositors contain the attack after different payment thresholds, corresponding to the sequential allocations composing the optimal mechanism designed by the bank, are reached. Containment requires the withdrawing line to be observable, which allows depositors to update their beliefs and infer the realized state of the economy. Being able to update their beliefs, agents play a series of signalling Bayesian coordination games after depositing, whose equilibria are defined by a subjective set of certainty equivalent constraints. The equilibria set of these games follows directly from the information and beliefs held by agents before depositing. That is, the set of games is played ex-ante, but agents consider all possible (subjective) expost realizations to define their strategies. If a self-contained run takes place, imposing suspension of convertibility would assign an inefficient allocation to a subset of depositors with probability one. However, and restricted to the theoretical framework defined within this paper, suspension constitutes an optimal containment policy if it is announced after
The literature on this subject is quite extensive. See Furman and Stiglitz (1998) for a summary of the main results.
7

agents deposit. On the other hand, the optimality of suspension does not necessarily hold if a more complex strategic dynamic environment with multiple banks is considered. We provide some basis for further research on this topic, since the current paper aims only to prove the existence of rational bank runs not based on the realization of an exogenous sunspot variable. The paper proceeds as follows. A description of the strategic environment derived from the optimization problem faced by depositors is given in section two. Section three introduces formally the pre and post-deposit coordination structures generated by the bank through its sequential mechanism and defines their set of equilibria, which allows for the existence of rational self-contained runs. Section four illustrates numerically the theoretical results presented in the previous sections. Section five deals with suspension of convertibility and proposes several extensions of the current model.

tion we refer to

f ( b ) = f ( b | b b ), b b i i i
Consider the following probability space (, I, F), where every state in the sample space, , determines how many and which agents are affected by the shock. Moreover, I is a -field defined on , and F is a probability measure on I. Define a random variable on the state space as follows
b : b {1 ,..., N } :
b where = b for some i = 1,...,# b . i The symbol # denotes the cardinality of the set defined after it, i.e. its number of elements. Note that while a given state determines how many and which agents are affected by the shock, the random variable reports only the former component. The number of agents affected by the shock is determined by the realization of the random variable and is Borel measurable with respect to the -field I, which allows both agents and the bank to infer the exact realization of the random variable, *, in period one upon observing the length of the withdrawing line.

2. Theoretical Strategic Behavior

2.1 INITIAL ASSUMPTIONS

The basic model follows Peck and Shell (2003), who define an identical environment to the original one described in Diamond and Dybvig (1983), except for the fact that a finite 8 set of identical agents, N, is assumed. This assumption allows for monitoring by depositors through a countable withdrawing line. There are three time periods, t=0,1,2. Each agent is endowed with one unit of an homogeneous good in period zero, which can be costlessly stored among periods. There exists a production technology that delivers one unit of output per unit invested if interrupted after one period, and R>1 units of output if the in9 vested unit is kept for two periods. Agents are subject to an exogenous shock in period one defined by the set of possible realizations of a given random variable, b, with an associated probability function f(b). To simplify notaThrough the paper N will also stand for the cardinality of the set. 9 We are assuming that agents can interrupt their investment after one period without suffering any penalty. Jacklin and Bhattacharya (1988) challenge this assumption in a framework with interim, as of period one, signals about the stochastic returns obtained in period two. This allows them to design rational, information-based runs dependent on the expected state of fundamentals defining the return.
8

The agents affected by the shock become impatient, or type 1, and value consumption in period one only. The remaining agents, which we refer to as patient or type 2, value coni sumption in both periods. Denote by ctk the amount of goods received by a type k agent in period t given state i. The state dependent utility of each agent is given by

i i i U ( c 11 , c 12 , c 22

i u ( c 11 ( b )) i if the agent is impatient ;b ) = i i b i b u(c 12 ( i ) + c 22 ( i ) ) if the agent is patient

where is the rate of time preference, b represents a given relative state of the econi omy, such that b b , defining the utility dei pendence on the realization of the random variable (as well as the sequential nature of the set of allocations offered by the bank), and u : + is increasing, twice continuously differentiable, and satisfies the Inada conditions, u'(0)= and u'( )=0. We assume the following functional form through the paper
u (c ) = c1 1

10

realization of the latter ones. Jointly with a sequential service constraint, the (assumed) altruistic bank maximizes the following value function V (b , f (b ))
max

i i i c 11 , c 12 , c 22 i = 1 ,..., # b

i =1 bb i

#b

j i ( bj bj 1 ) u ( c 11 ( bj )) + f (b ) i b i b i b j = 1 + ( N i ) u ( c 12 ( i ) + c 22 ( i ))

with > 1 everywhere. It is generally assumed that before a financial intermediary is created, agents maximize expected utility conditional on their information set, a = {N , b , f (b ), Ta } , where the subscript refers to agent a, and Ta stands for the realization of the shock corresponding to agent a. The standard assumption regarding the information sets of potential depositors is that all of them have identical ones. Thus, if a financial intermediary is formed, it should inherit the homogeneous information set of its depositors regarding the expected state of the system, b and f (b ) . That is, the potential bank and its depositors share information sets, as well as beliefs, except for the shock defining the type, which is privately observed by each agent and cannot be verified by the bank, assumed to be the only 11 provider of liquidity in the economy. The main results obtained in this paper derive from relaxing the assumption assigning identical information sets to the bank and its deposi12 tors.
2.2 CONTRACT-INDUCED COORDINATION GAMES

where # b denotes the cardinality of the set of possible realizations, b = 0 , and (bj bj 1 ) repre0 sents the number of agents contained between two consecutive realizations within b , subject to the set of budget constraints, which must be satisfied b b i
i

i c i (b ) + ( N b ) c 12 ( b ) + 22 i i i R i = 1,..., # b

j =1

b j

j bj 1 ) u ( c 11 ( bj )) +

= N,

and a corresponding mechanism compatibility condition. To highlight the difference between the standard models studied in the literature and the current setting, we present the incentive compatibility constraint, icc henceforth, 13 defining the optimal mechanism of the bank

i =1

# b

i f b ( b ) u ( c 22 ( b )) i i

i =1

# b

( bj bj 1 ) j u ( c11 ( bj )) + i b + 1 i , (1) f b (b ) i 1 j =1 i+ u ( c11 1 ( b + 1)) + b i i + 1

such that Consider the optimization problem faced by an altruistic bank, given the previous framework and conditional on its information set, b = {N , b , f ( b )} , which is assumed (by the bank) to be identical to the one of its depositors, a , except for the privately observed type
10 11

f b (b ) = i

N 1

[1 (
'= 0

b i

/ N ) f ( b ) i

[1 ( ' / N )] f ( ' )

b , ' b i

For simplicity, it is generally assumed that =1. Jacklin (1987) provides a formal justification regarding the inefficiency caused by alternative asset markets introduced within the current context. In short, Jacklin (1987) shows that the demand deposit contracts designed by the bank achieve greater risk sharing than equity shares if trade is restricted among agents. 12 The lack of financial transparency was presented by Furman and Stiglitz (1998) as one of the main plausible causes triggering the Asian crises.

where f (b ) stands for the ex ante (prior) i probability assigned to the number of impatient agents being b , with b b {1,..., N } . This i i function is generally assumed to be common knowledge among all agents, and therefore, also between depositors and the bank, in period zero. Patient agents update this probabi13

Obviously, non-negativity constraints are also imposed on all consumption allocations.

10

lity in period one, after receiving the type14 determining signal, using Bayes' rule. The updating process delivers the subjective probability, as of each agent, of having b impatient i depositors in the system, conditional on the agent being patient, f b (b ) . These probabilities i are used by depositors to calculate the expected payoffs on which to base their strategic behavior in the post-deposit game generated by the bank contract structure that solves the previous optimization problem. Beliefs are a direct function of f (b ) , assumed to be i uniquely defined with the probability space and known by all agents. This homogeneity assumption implies that depositors share the beliefs of the bank, denoted by f b (b ) , the subi script b representing the bank, when defining their optimal strategies. Compare now the incentive compatibility equilibrium condition to the set of conditions required to guarantee the existence of stable equilibria through the set of post-deposit games defined for each w b , where #
b

w w E[u (c 22 (w ))] u (c11+1 (w + 1)).

(2)

These equilibrium constraints imply that patient agents prefer to misrepresent their type if the utility derived from the allocation offered to the next impatient agent in line is strictly larger than the expected utility from waiting to withdraw until period two. That is, patient agents attack the bank if the allocation received by impatient depositors is larger than the certainty equivalent corresponding to the lottery faced by waiting. The set of lotteries from which the cec are derived is based on the set of realizations observed by depositors and their subjective beliefs, updated following Bayes' rule conditional on the number of withdrawing agents in line. Upon direct observation of both post-deposit equilibrium stability requirements it is clear that the set of cec imposes a stricter constraint on the mechanism than the icc. Thus, the stability conditions (at least a subset of them) defined by the set of cec would be violated if the icc binds in equilibrium with equality. It should be emphasized that the entire optimization problem is based on the set of expected realizations of the random variable and its induced probability distribution. The sequential structure of the service constraint defining the solution mechanism is directly generated by these variables through the optimization problem faced by the bank. Moreover, the equilibrium of the model is determined by the effects of b and f (b ) on the set of cec. These constraints do not only allow for the implementability of the mechanism obtained as an optimal solution to the optimization problem, but define its stability. As we will show, the set of cec guarantees the existence of no-run equilibria within the set of Bayesian games generated by the mechanism, such that we can rely on the revelation principle to eliminate any possible attack from the optimal communication strategies of depositors.
2.3 UPDATING THE BAYESIAN BELIEFS OF DEPOSITORS

stands for the number of withdrawing agents observed in line and b# corresponds to
w
b

the supremum of .
b

The icc condition is based on the expected payoffs that the bank assumes its depositors calculate if they are not able to observe the length of the line before deciding whether or not to withdraw, leading to a static (calculated and fixed as of period zero) inequality in ex15 pected payoffs
i i+ E[u (c 22 (b ))] E[u (c111 (b + 1))], i i

(1)

which defines a unique Bayesian post-deposit game and its corresponding equilibrium. The ability to observe the length of the withdrawing line leads depositors to (subjectively) generate a set of post-deposit games for each w b , as well as a set of certainty equivalent # constraints, cec hereafter, defining the respective equilibria
b

14

Peck and Shell (2003) were the first ones to define explicitly the belief formation process of the agents. Endowing agents with the ability to generate subjective expected payoffs is essential to model strategic interactions among depositors in the postdeposit game. 15 It is a known result that the optimal demand deposit contract assigns c i ( b ) = c i ( b ) = 0 , b b .
21 i 12 i i

Assume from now on that the prior set of expected realizations, b , and the probability function defined on it, f (b ) , are no longer common knowledge in period zero and may differ between the bank and its depositors. This restriction allows for heterogeneous be-

11

liefs among agents, caused by differences between the set of expected realizations (and its associated probability function) defining the optimal mechanism and the set of possible 16 realizations considered by depositors. The latter set equals either {1,...,N}, if agents have no information about the possible realizations of , i.e. agents exhibit total aggregate uncertainty, or a , if information differentials lead to different expected sets of realizations. We will restrict attention to a information sets generated by mean preserving spreads on the set of expected realizations observed by the bank. Such a restriction represents a natural economic framework to analyze information differentials. This is not equivalent to affirm that this restriction is either necessary or sufficient to generate rational self-contained runs. Finally, assume that, given the informational advantage that the bank may develop over its depositors, b and f (b ) define the correct expected state of the world. Consider the set of realizations expected by the agents before depositing, a , and its associated subjective probability function f a (a ) . Define f (ia | p, w ) as the updated Bayesian beliefs of depositors conditional on the agent being patient, p, and the number of withdrawing agents observed, w ,
f (ia | p , w ) =

signed by the agents to the set of possible states, f a (a ) . Beliefs are updated for each value of w = 1,..., a , # with a representing the supremum of the set #
a

. From the point of view of a given depositor, the length of the line represents the communication strategies of the remaining agents, since we are considering equilibria in pure strategies and actions are equivalent to messages used as noiseless signal of the state. Without a sunspot variable coordinating all depositors on a run, withdrawals behave as public signals reflecting the strategies of depositors. An observed signal, in the form of w , modifies the beliefs of agents regarding the state of the world conditional on a a . Indii vidual agents assume all remaining depositors to share their set of beliefs and act accordingly. Thus, agents expect other depositors to misrepresent their types when there exists an incentive to do so themselves. Agents are, at the same time, aware of the fact that by attacking the bank they will be modifying the final set of realizations, and the behavior of other depositors who react to their signal. The subjectively updated Bayesian probability function of depositors must reflect all the described information properties of the withdrawing line.
a

f (ia | w ) f ( p | ia )
N 1 '=1

f ( p | ' ) f ( ' | )
w

' a

The term f ( p | ia ) measures the subjective de17 gree of pessimism inherent to each agent. We assume a neutral distribution through the paper, but the model can easily be extended to analyze the effects of pessimism, and for that matter optimism, on the deposit and withdrawal decisions of agents.

with an updated subjective probability for each agent defined by

f ( ia | w ) =

f ( w | ia ) f a ( ia )
N

'= 1

f ( w | ') fa ( ')

' a

where a a . The probability f (ia | w ) reprei sents the private beliefs of depositors about the possible states of the world defined within their subjective set of realizations, a a , after i observing w agents in line. These beliefs are directly based on the probability function as16 17

Neutrality of the distribution is usually assumed under risk,

f ( p | ia ) = [1 (ia / N )] . However, in an uncertainty context,

associated with a subjective probability function f u () , prospect theory suggests pessimistic biases in the subjective expectations of agents, f u ( p | ia ) < f ( p | ia ) , for a < * , with * [0, N ] i defined by each depositor. In words, agents overvalue the subjective probability of being affected by the shock in relatively good states of the world, i.e. realizations of the random variable below a given cut off point , and undervalue it in the bad ones. For additional prospect theory results and a quantitative analysis of the subjective parameters see Kahneman and Tversky (1979), and Tversky and Kahneman(1992).
*

Heterogeneous expectations are justified as a direct consequence of information differentials between the bank and its depositors. Jacklin and Bhattacharya (1988) rely on differences in the risk aversion coefficient of agents to justify heterogeneity in the acquisition of information. We could also assume that financial institutions have easier access to information sources due to scale economies reducing the acquisition costs of information relative to individual agents, or that they have private incentives, either altruistic or egoistic, to develop an informational advantage over their depositors.

12

3. Post-Deposit Coordination Games

3.1 PRE AND POST-DEPOSIT EQUILIBRIA

runs out of reserves. The remaining allocations composing the mechanism vector, i (c2 (b )), i = 1,2 , define the consumption given to i patient agents in period two, depending on the number of withdrawing agents in period one. The strategies of depositors, based on their subjective expected payoffs, defining the Nash equilibria of the set of post-deposit games generated by the bank through m( f (b )) , depend on w . That is, a signalling Bayesian game will be played by all patient depositors for each w a . The post-deposit game theoretical set# ting induced by the mechanism is no longer unique, as is the general case considered in the literature, but composed by a set of signalling Bayesian coordination games. Denote this set by . The dependence of the expected payoffs of depositors, defining each game within , on the value of w must be reflected in the payoff matrix of the corresponding game, denoted by (w ) , which is given by
a

19

Despite the existence of two differentiated coordination games, recognized by Adao and Temzelides (1998), and Peck and Shell (2003), the unique equilibrium (in pure strategies) defined by the depositing decision of agents is assumed to pervade through both games, unless altered by an exogenous sunspot shock in the interim period. There exists a tacit agreement regarding the unique run-free Nash equilibrium of the pre-deposit coordination game. Agents do not deposit funds if they expect a run in the interim period, see Postlewaite and Vives (1987), unless it is caused by a sunspot variable taking place with low enough probability. The post-deposit game is therefore identical to the pre-deposit one, and preserves the same run-free equilibrium. This result relies heavily on the assumed homogeneity of beliefs (and strategies) among depositors, and, particularly, between depositors and the bank. Once information and beliefs are allowed to differ, the resulting equilibrium supports rational self-contained runs expected by the agents before depositing, not necessarily generated by an exogenous sunspot variable.
3.2 THE SET OF POST-DEPOSIT GAMES

The solution to the bank optimization problem is given by a mechanism, m( f (b )) , consisting of a vector of optimal and feasible state contingent allocations defined for each value in b the set b = {1 , b } , and their corresponding 2 18 b probabilities f (b ) = { f (1 ), f (b )} 2
1 b b 2 m( f (b )) = (c1 (1 ), c12 (b ), c1 (1 ), c 2 (b )) 2 2 2
b where c11 (1 ) denotes the consumption offered b to the first 1 agents in line who declare being impatient, while c12 (b ) is allocated to each of 2 the remaining impatient agents in line, in case of an attack up to the point where the bank

19

18

hold for any n-dimensional, n N , set of realizations, as long as the mechanism allows for the existence of runs on the bank. At the same time, since optimality requires i i c21 (b ) = c12 (b ) = 0, b b , subscripts will denote from i i i now on both the type of depositor and consumption period.

The dimension of has been chosen to simplify the presentation without loss of generality. The main results of the paper
b

The problem presented by suspension of convertibility and other alternative run-proof mechanism design policies is not a simple one. Their effect on the final depositing decision of agents depends on the type of information friction generating the biased beliefs of depositors. Announcing suspension before agents deposit becomes optimal with biased subjective probabilities on a unique sunspot variable realization, but it is not necessarily optimal if beliefs are generated through a mean preserving spread on the expected set of realizations. See Section 5 for a more detailed analysis of this topic.

13

-(i-w)
withdraw withdraw not withdraw
w r 1 w

i
where

E[u (c ( ))], E[u (c ( ))]

r 1 r u (c2 ), E[u (c1r 1 (w ))]

u (c1p (w + 1)), E[u (c2p (w + 1))]

nr nr E[u (c2 (w ))], E[u (c2 (w ))]

not withdraw

b b N 1 1 2 b 1 w 1 b u (c1 (1 )) + N E[u (c1r (w ))] = N w u (c1 (2 )) w 1 b 1 b b + N u ( N 1 c1 (1 ) (N 1 1 )c12 (b )) 2 w r b 1 b b u (c2 ) = u ([ N 1 c1 (1 ) (N 1 1 )c12 (b )]R) 2

b N 1 1 2 b b w 1 b u (c1 (1 )) + N 1 E[u (c1r 1 (w ))] = N11 w u (c1 (2 )) w

1 b b u (c1 (1 )) if w < 1 u (c ( + 1)) = 2 b b u (c1 (2 )) if w 1 p 1 w

E[u (c2p (w + 1))] =

b ia 1

b 1 ( N ia c1 (1 )) R f (ia | p, w + 1)u N ia b 1 b b ( N 1 c1 (1 ) (ia 1 )c12 (b )) R 2 f (ia | p, w + 1)u a N i 1 b ( N ia c1 (1 )) R f (ia | p, w )u N ia b 1 b b ( N 1 c1 (1 ) (ia 1 )c12 (b )) R 2 f (ia | p, w )u N ia

b ia >1

nr E[u (c2 (w ))] =

b ia 1

b ia > 1

E[u (c1r (w ))] corresponds to the expected

payoff, in utility terms, received by an agent if all depositors attack the bank in period one. The superscript in N indicates the number of agents withdrawing funds in period one, i.e. the length of the withdrawing line, when no subscript is used to define the variable. Given

place after observing w agents in line. All agents are assumed identical in this respect. There is no agent with an exogenous a priori advantage allowing him to reach the line before others do.
r u (c2 ) derives directly from the budget con-

N and b ,

b N 1 1 refers to the probability N w

of reaching the line within the interval b (1 , N 1 ] if a general run on the bank takes

straint of the bank, and represents the utility value of consumption for the last patient agent if all other depositors have declared to be impatient.

14

Note that if
N

1 = 1

>

a patient depositor who knows to be the last agent in line has no incentive to misrepresent her type. The last depositor, if patient, receives in period two the return of whatever remains after all the other agents have withdrawn invested in the long run technology. is the utility obtained by a patient depositor who withdraws funds in period one after observing a line of length w , while the remaining patient agents do not misrepresent their types simultaneously. and E[u (c2p (w + 1))] follow from the set of cec conditions defining the no-run equilibria of all possible post-deposit games played 20 by patient depositors
nr E[u(c2 (w ))] u(c1p (w + 1)), w aa . (2) #
nr E[u (c2 (w ))]

u (c1p (w + 1))

each value of w a . Agents are able to calcu# late in period zero, before knowing their type, the set of all possible games they will be facing in period one depending on the number of withdrawing agents in line. This is the case since both the withdrawing and depositing decisions are based on the expected set of equilibria subjectively obtained by the agents in period zero for each (w ) , with w = 1,..., a . Formally, every agent defines her # set of strategies based on the subjective expected utilities obtained from the mechanism designed by the bank for all possible values of
a a

w a #

E[ua ] : M {1,...,aa } Fa (a ) # {E[u(m a ( f a (a ) | m( f (b )),w ))] : w aa )} #

where M corresponds to the set of mechanisms offered by the bank, with m( f (b )) M , while Fa (a ) is the set of subjective probability functions defined by the agents on their expected set of realizations, f a (a ) Fa (a ) . The previous function is a mapping from the set of allocations offered by the bank and the set of subjective beliefs held by agents to the set of expected utilities derived from the subjective mechanisms calculated by the depositors, E[u (m a ( f a (a ) | m ( f (b )), w ))] , that are used to define the entries of the (w ) payoff matrix, for each w a . Every subjective mechanism #
a

Differentiating rational self-contained runs from sunspot-based general ones requires the run equilibria of all games, (w ) with w = 1,..., a , to be defined on a unilateral basis # and not to involve the entire subset of patient depositors. We want to prevent the generation of herds and allow for agents to distribute the attacks depending on the value of w . Thus, the set of run equilibria must consist of selfcontained attacks, based on nr u (c1p (w + 1)) > E[u (c2 (w ))], that do not spread to the entire subset of patient depositors, which is the case as long as
a

calculated for a given realization of w , m a ( f a (a ) | m ( f (b )), w ) , defines a corresponding expected game, (w ) , within the set . Thus, the set of games played by the agents is the image of a function of all pairs of the form (m a ( f a (a ) | m( f (b )), w ), f a ( a )) .

E[u (c2p (w + 1))] > E[u (c1r (w ))], w aa . #

3.3 COMMUNICATION STRATEGIES

This section introduces formally the communication strategies of patient depositors defining the possible equilibria of each game within the set of post-deposit games generated by
20

The bank, endowed with the information set b = {N , b , f ( b )} , designs an optimal incentive compatible mechanism m( f (b )) , using the following allocation rule

: F ( b ) [ C ] 2 (#

: f (b )

i m ( f ( b )) = ( c k ( b )) i

i = 1 ,..., # b k =1,2

When

defining

nr E[u (c2 (w ))] ,

E[u (c2p (w + 1))] ,

r E[u (c1r (w ))] and u (c2 ) , we have assumed that all impatient a agents up to ( N 1) are paid c12 (b ) , and that the last 2 #
a

remaining depositor receives a strictly positive consumption allocation in case of a general run. While this is not necessarily true, it has been done to simplify the matrix entrances and does not affect the results presented.

where F (b ) is the set of probability functions that may be induced by a given set of realizations, b . This rule results in the set of optimal (and feasible) state contingent allocations that will be used to define the mechanism.

15

Note that the domains of the allocation rule and the optimal mechanism calculated by the bank have been defined in terms of F (b ) . This dependence highlights the stochastic nature of the optimal allocation set, the mechanism, and the resulting set of post-deposit games, which are all based on the expectations hold by the bank, but not necessarily by its depositors. The communication strategy of a patient depositor, a , defined for each (w ) , with
w a #
a

3.4 POST-DEPOSIT EQUILIBRIA SET

In order to study the interim period general rational expectations equilibrium, we must consider the set of equilibria defined for each and every Bayesian game played by patient agents after depositing, (w ) , for any possible value of w {1,..., N } . Definition 3.1 Bayesian Equilibrium of a Post-deposit Game A perfect Bayesian equilibrium of a signalling interim (post-deposit) game, ( ) , is given by a communication strategy, ( ) , and the corresponding updated beliefs f ( | p , ) , such that a N \ [ ] , where N defines the set of all agents, and [ ] , j = 1 ,..., # , stands for the set of agents affected by the shock, and for all
w
* a w a i w
b j b j

, is given by the following map

a : {w, nw}, a patient

that assigns to each game (w ) within an equilibrium withdrawing decision, consisting of either withdrawing funds, w, or waiting, nw. To simplify notation, we will denote the communication strategy corresponding to the a-th depositor and game (w ) by a (w ) . The communication profile of a patient depositor is given by a vector of messages (communication strategies) defined for each one of the games expected to be played in period one
a a ( ) = ( a ( 1 ),..., a ( a )) { w , nw } #
a

w = 1 ,..., a #

# a

For the entire post-deposit game theoretical structure to hold, there must exist at least one Bayesian Nash equilibrium in pure strategies per game, defined by the optimal communication strategies of patient depositors after observing w agents in line and updating their subjective beliefs.

a patient

The vector of messages derived from the communication strategies of all depositors is simply w , since the message space is equivalent to the space of actions. Agents consider a set of optimal communication profiles formed by their optimal vector of messages and those of all remaining depositors
* * { a (), a ()} = a , a patient
* where a () assigns an identical optimal communication profile to all patient depositors, given by a* () . This assumption is justified by the lack of communication among depositors, standard to this type of models, and the absence of any additional signals privately received by depositors regarding the beliefs of other agents, a la Morris and Shim, in the interim period. The union set of optimal communication profiles defining the final value of w is given by , and includes the vectors of messages generated by all depositors, allowing for heterogeneous beliefs among them.

16

* E[u(c(w ) | (w ),a (w ))] E[u(c(w ) | (w ),a (w ))]

which in expected utility terms translates into

iaa

f ( f (

a i a i

* | p, w )ua (m a ( f a (a )) | m( f (b )), w , a (w ))

iaa

| p, w )ua (m a ( f a (a )) | m( f (b )), w , a (w ))
that runs on the bank can be prevented, except if caused by a sunspot realization, as long as each and every post-deposit game played by patient depositors has a no-run Nash equilibrium where agents can coordinate their communication strategies. We define now the set of rational expectations equilibria induced by m( f (b )) . In order to do so, the Bayesian equilibrium definition presented above must be integrated in a setting with multiple games, whose payoffs matrices are defined as functions of the sequential allocations composing the mechanism, and the set of subjective beliefs held by depositors. Definition 3.3 Rational Expectations Equi-

Runs are eliminated from the communication strategies of depositors if, and only if, we are able to guarantee the existence of a non-run Bayesian Nash equilibrium for each and every post-deposit game expected to be played by the agents. This stability condition follows directly from the application of the revelation principle to the entire set of interim games. Definition 3.2 Revelation Principle Define

the set of messages sent by the agents to be the set of withdrawing depositors, w , due to the existing equivalence between messages and actions. When an agent withdraws funds is also sending an observable message revealing that is impatient. The message space is then {w, nw}N , with w {w, nw}N , and is defined by the communication strategies of depositors for each and every possible value of w . Consider the stochastic allocation rule defined by the bank depending on the set of expected messages, ( f (b )) . Assume that a mechanism is given, m( f (b )) , that both induces and is defined by the expected set of messages b , with an associated allocation function ( f (b )) , such that the set of Bayesian post-deposit games generated by m( f (b )) has a set of Bayesian Nash equilibria defined by the communication profile * * * (w ) = (1* (w ), 2 (w ),..., N (w )) , for all w b , # where all patient agents report truthfully. That is, for a given realization of the random variable defining the true state of the economy * b* b , we have w = b* , and a (w ) = nw , w b* a N \ [b* ] , and .
b*

libria of the Deposit Games A set of rational expectations equilibria for the corresponding set of deposit games, both pre and post, is given by an implementable mechanism designed by the bank, m( f (b )) , defining a set of allocation vectors for all possible realizations within b , and a communication profile, * (w ) , w a , generated by the communica# tion strategies of patient depositors, such that
a

The revelation principle states that, given the previous Bayesian equilibrium, there exists a direct revelation mechanism, ( ' (w ), ' ) , with ' = ( f (b )) , since it is a stochastic rule, such that ' (w ) = * (w ) , and all the agents accepting the mechanism report truthfully.
In other words, the revelation principle states

(i) The set of allocation vectors composing the mechanism maximizes the expected utility of depositors, m* ( f (b )) arg max V (b , f (b )) , and satisfies the budget, certainty equivalent (incentive compatibility), and non-negativity constraints. V (b , f (b )) is the true value function, defining the optimization problem faced by agents in period zero for the probability function, f (b ) , and the set of expected realizations, b , observed by the bank, which correspond to the true expected state of the economy. (ii) The mechanism satisfies a sequential service constraint. The consumption allocation received by an impatient agent in period one, c1 (w ) , depends only on the number of agents standing before him in line, w 1 . The num-

17

ber of allocations defining the sequential withdrawing intervals of the mechanism, c1i (b ), b b , is given by the cardinality of the i i set b . (iii) The set of optimal communication profiles, , based on the private beliefs of depositors for a given (optimal) mechanism m* ( f (b )) , with
* * a ( a (1 ),..., a (aa )) a , a patient #

where (a* (1a ),..., a* (a# )) represents the vector of optimal communication strategies for agent a given the set of games generated by all the possible values of w that can be observed according to her private set of expected realizations, a , defines a set of perfect Bayesian equilibria for the set of signalling games played in the interim period, a N \ [b* ] , and w {1,..., b } . The homogeneity assumption on # the set of private beliefs defined among patient agents, Fa (a ) , simplifies this requirement to a unique communication profile, since , for all depositors a a' , with a ( ) = a ' ( ) a, a ' N \ [b* ] .
a
b

patient depositors through m* ( f (b )) , and w f (ia | p, w ) , ia a , after observing withdrawing agents. Given the cec subjectively calculated by depositor for (w ) , a* (w ) = nw , a N \ [b* ] , if there exists a Bayesian Nash equilibrium in this game defined by the truthful communication strategies of all depositors. However, if u (c1p (w + 1)) > E[u (c2nr (w ))] , a* (w ) = w , a N \ [b* ] . In this case, the perfect Bayesian equilibrium of the signaling game (w ) is given by either an unilateral attack, if E[u (c2p (w + 1))] E[u (c1r (w ))] , or a general run on the bank, if E[u (c2p (w + 1))] < E[u (c1r (w ))] . The revelation principle does not hold and cannot guarantee a no run (truth-telling) equilibrium for any subset of post-deposit games where nr u (c1p (w + 1)) > E[u (c2 (w ))] . (vi) Given m* ( f (b )) , the set of subjectively updated beliefs, f (ia | p, w ) , for all a a , geni erating the set of optimal communication profiles for all patient depositors, $\Phi$, and the set of Bayesian Nash equilibria defined by the optimal communication profiles (based on a and f a (a ) ) for each and every post-deposit game within the set , with (w ) , w {1,..., a } , agents prefer depositing funds in # the bank to autarky.
a

(iv) Agents exhibit Bayesian Rationality when updating the set of subjective beliefs used to define their communication strategies in each one of the corresponding post-deposit games, (w ) , with w {1,..., a } . The set of optimal # communication profiles, , gives place to a set of noiseless signals represented by the length of the withdrawing line, w . These signals are used by patient depositors to update their beliefs about the expected state realization and the corresponding consumption received in period two, which derives directly from the set of equilibria generated by the optimal communication profiles within through the expected remaining games in the set, (w ) , for w {1,..., a# } . Subjective beliefs generate the forward induced structure that defines the set of optimal communication profiles of patient depositors, giving place to the set of Bayesian Nash equilibria for the entire set of interim games.
a a

The literature on bank runs considers a unique post-deposit game for a given f (b ) , whose norun equilibrium is guaranteed by a direct application of the revelation principle. The equilibrium set defined for the subjective set of signalling Bayesian games allows for bank runs that follow from the optimal communication strategies of patient depositors.
3.5 MECHANISM MONOTONICITY AND SELFCONTAINED RUN EQUILIBRIA

(v) The revelation principle holds when active throughout the entire set of post-deposit games. Consider the game (w) , defined by

Sunspot variables lead to general runs on the bank if realized, which conditions the depositing decision of agents on the imposition of an exogenous constraint on their realization probabilities. On the other hand, given f a (a ) , bank runs must self-contain to be compatible with the rational deposit equilibria definition presented above and for agents to deposit in period zero. A run on the bank can be contained if the mechanism allows for individual incentives to run for a given subset of values

18

defined within the sequential intervals induced by b , but at the same time keeps all patient depositors from withdrawing simultaneously, given f a (a ) and assuming that all agents share identical beliefs. That is, runs are contained if, after having an incentive to unilaterally misrepresent their types, the allocation adjustment process defined by the mechanism suffices to coordinate the strategies of patient depositors back on the truth telling equilibrium. We will make use of the following lemma to illustrate the run equilibrium results presented in this section Lemma 3.4 Assume that a general run on the bank defines one of the possible equilibria for the set of post-deposit games, . That is, there exists a value of w smaller than N, denoted by r , such that the bank does not have enough resources to honor the sequential allocations of the mechanism in case of a run
[ N i = 1 (
# b b i b i 1

which is identical to the equilibrium condition defined for a unary set. From the assumed concavity of u(c) and the value of R(>1) we obtain
# c2 (b b ) > c1# (b b ) . # #
b b

Therefore, the consumption allocations received by patient depositors for values of w > b 1 and a given c ( ) decrease mono#
b

# b 1

b # b

tonically in w , since
[N

# i=1
b

(
b #

b i
b

b i1

) c 1i (
b

b i

) [( N 1 )
b #
b

b #

] c 1# (

)] R < c 1# (
# b 2 b # b # b 1 2

which implies that c ( ) < c ( ) . A simple backward induction argument applied on the remaining equilibrium conditions delivers the required result. The first order optimality condition defining the allocation that corresponds to the previous element within the realization set, b# 1 , is given by
b # b 1
b

)c ( ) (
i 1 b i r

b # b

)c

# b 1

b # b

)] R

N c 1# ( b b ) #
b

<

If none of the cec within the set used to define the optimal mechanism binds in equilibrium,21 all the elements composing the set of optimal allocations offered to both types of depositors, i (c1i (b ), c2 (b )), b b , are monotonically dei i i creasing in b . i
Proof. The proof for a unary realization set, b {1 } , is presented in Green and Lin (2003). They illustrate the increasing monotonicity of 1 b b c1 (1 ) and c1 (1 ) relative to decreases in the 2 b value of 1 . Their proof derives directly from the assumption of a coefficient of relative risk aversion larger than one. The interested reader is referred to Lemma 1 in their paper. In order to generalize this result note that, for any cardinality of b , if none of the cec binds in equilibrium, the optimization problem defining the mechanism gives place to the following first order condition for b#
b

# u ' (c1# (b b )) = Ru ' (c2 (b b )) , # #

21

This assumption is made to simplify the presentation, and follows directly from the simulation results.

19

u ' (c
implying that

# b 1 1

b # b 1

)) =

# # R[ f (b b 1 )u ' (c2 1 (b b 1 )) + f (b b )u ' (c2 (b b ))] # # # #

f (b b 1 ) + f (b b ) # #

u ' (c1# 1 (b b 1 )) < u ' (c1# (b b )) , # #

and, therefore,

# b 1 1

b # b 1

)>c

# b 1

b # b

).

It follows directly from this last inequality that, given c1# 1 (b# 1 ) , the allocation offered to patient depositors within the interval ( b 2 , b 1 ] is monotonically decreasing in # #
b b
b b

w . Thus, we must have

# # c2 1 (b b 1 ) < c2 #
b b

for any h <# b , after observing w (b1 , b ] i i agents in line, provides patient depositors with an incentive to unilaterally misrepresent their types, as long as the second inequality is not reversed by a posterior allocation, but does not necessarily degenerate in an immediate general run on the bank. The following payoff matrix represents the strategic environment on which the previous inequality is based

(b b 2 ) . #

-(i-w)

The same reasoning applies to all the remaining allocations within the mechanism, offered to both patient and impatient depositors. This completes the proof. The introduction of a dynamic updating process, based on w , in the formation of expected payoffs defined by patient depositors together with the decreasing monotonicity of c1i (b ), b b , lead to the following result i i Theorem 3.5 Assume that the bank does not have enough resources to honor the sequential allocations composing the mechanism in case of a general run. If a mechanism is vulnerable to general runs, there exist subjective sets of random variable realizations, a , with their associated probability functions, f a (a ) , leading to self-contained attacks on the bank. Proof. The monotonic and optimality qualities of (c1i (b ), c2i (b )), b b , jointly with the exisi i i tence of r , imply that there must exist at least one value p < r satisfying
c
#b 1

withdraw withdraw

not withdraw

E[c ( )],
r 1 w

c1p (w + 1), E[c2p (w + 1)]

nr E [ c 2 ( w )], nr E [ c 2 ( w )]

i
not withdraw

E[c1r (w )]
r c2 ,

E [ c1r 1 ( w )]

Consider the northeast corner of the matrix. Unilateral attacks take place if p w nr w , and continue as long as c1 ( + 1) > E[c2 ( )]

E[c2p (w + j)] < c1p (w + j +1), with N2 w j > 1

The (Bayesian) updated private beliefs hold by depositors for w (b1 , b ] define a selfi i contained run on the bank, that does not extend to the entire set of patient depositors, conditional on E[c2p (w + 1)] E[c1r (w )] . This stability condition guarantees the containment of general runs uniquely within the interval (b1 , b ] . The self-contained quality of bank i i runs requires that no general run is triggered through any of the games generated by the set of allocations defining the mechanism. That is, given the private beliefs of depositors defining their expected payoffs through the entire set of games, the following condition must be satisfied w b#
b

( )
b #b

[ N i =1 (b b1 )c1i (b ) [ p b b ]c1# (b b )]R i i i # #

#b
b

N p

<

c1h (b ) h

with h <# b , and p a . Thus, a subjective probability distribution leading the beliefs of depositors to satisfy

E[c ( + 1 | f a (a ))] E[c1r (w )] ,

p 2 w

20

where E[c2p (w + 1 | f a (a ))] represents the expected payoff received by a patient agent in period two if an additional withdrawal takes place after observing w agents in line. When calculating this value agents must consider the cumulative effect that unilateral withdrawals, made by patient depositors through the mechanism intervals prior to the realization of w , have on the expected payoffs of the remaining depositors. The general containment condition defined above derives in a natural way from the game theoretical structure generated by i b i b b b . To see why this is the (c1 (i ), c2 (i )), i case, take the payoff matrix defined by the lowest value of w giving place to unilateral withdrawal incentives among patient depositors -(i-w)
withdraw withdraw E [ c1r ( w )], not withdraw c 1p ( w + 1),

For a run to take place and, at the same time, remain self-contained within the interval (b , b+1 ] , there must exist a value aj a , such i i that the following relations are satisfied

aj (b , b+1 ] , i i
E[c2p (w + 1 | f a (a ))] E[c1r (w )], w (aj , aj+1 ]
Assume that w + 1 corresponds to b+1 and the i allocation adjustment of the mechanism for c1p (w + 2) (b+1 , b+2 ] is large enough to eliminate i i any unilateral incentive to withdraw funds

E[c2p (w + 1 | f a (a ))] c1p (w + 2) > E[c1r (w + 1)]

If the attack is contained at w + 1 , depositors do not expect a general run on the bank when playing (w + 1) unless it is triggered by a sunspot. The general run containment conditions defined for each sequential interval composing the mechanism lead to the final cumulative containment condition
u ( c 1# ( b b )) < #
b

E [ c1r ( w )]

E [ c 2p ( w + 1 | f a ( a ))]
nr E [ c 2 ( w | f a ( a ))], nr E [ c 2 ( w | f a ( a ))]

not withdraw

r c2 ,

E [ c1r 1 ( w )]

Consider again the northeast corner of the matrix. Unilateral attacks by patient depositors take place if c1p (w + 1) > E[c2nr (w | f a (a ))] , and persist through w + 1 if

' a

[ N w' + # p c 1 ( w ) ] R =1 f ( ' b b | p , w ) u # N ( '+ # p )

E[c2p (w + 1 | f a (a ))] < c1p (w + 2) ,

as follows directly from simple observation of the matrix defined below. At the same time, the private set of beliefs, f a (a ) , triggers a run that does not extend to the remaining patient depositors if E[c2p (w + 1 | f a (a ))] E[c1r (w )] . The payoff matrix generated after the unilateral attack of a patient depositor is based on
E[c2p (w + 1 | f a (a ))]

In this case, after agents have considered the cumulative effect of any possible withdrawals made by patient depositors within b , denoted by #p, there are no unilateral incentives among the remaining patient depositors to run on the bank.

4. Main Results
There are no a priori restrictions on the private information sets acquired by agents in period zero. However, we can always define sets that serve exactly our purpose of generating selfcontained runs, without imposing any criteria on how information is obtained. The generic quality of the model described allows for the simulation of any agent type fitting precisely our required characteristics. This would present us with a justification problem from an economic perspective. The mean preserving spread restriction on the information sets of depositors will provide us with such intuitive

-(i-(w+1))
withdraw withdraw not withdraw E [ c1r ( w + 1)], c1p ( w + 2),
E [ c1r ( w + 1)] E[c 2p ( w + 2 | f a ( a ))]

i
not withdraw

r c2 ,

E [c 2p ( w + 1 | f a ( a ))],

E[c1r 1 ( w + 1)] E [c 2p ( w + 1 | f a ( a ))]

21

requirement. The less transparent the financial system considered or, equivalently, the higher the information acquisition costs, the larger the information dispersion suffered by depositors, and, therefore, the weaker and more runprone the financial structure of the system 22 becomes. Communication is not allowed among agents, as is generally assumed in this type of models. However, allowing for agents to communicate does not necessarily eliminate the incentives of depositors to attack the bank. All agents expecting a self-contained run have a clear incentive to report the set b (together with f (b ) ) as the true expected one. This can be interpreted as an attempt to gain withdrawing advantage by decreasing the expected number of depositors who attack the bank. Adding new levels of uncertainty to the current framework would allow for a more complete, and complex, analysis of the strategic interactions among depositors and their effect on the generation of runs. The self-fulfilling nature of the attacks does not contradict the rationality assumption imposed on the expectations formation process of depositors. Runs take place before the state of the economy is realized, which prevents them from being noticed by either the bank or its depositors until withdrawals stop. We are applying Adam Smith's invisible hand principle to the depositing decision of agents. Depositors do not consider the real nature of the bank as long as they are able to monitor the withdrawing line and guarantee that the bank honors the mechanism offered, despite knowing that it is not the optimal one according to their information sets. If agents deposit funds and agree to the terms of the mechanism, it is because they strictly prefer the expected utility derived from depositing to an autarky situation. The mechanism does not transmit information about the true expected state of the economy, since depositors do not know if it has been designed by an altruistic bank. At the same time, agents do not care about this fact, as they are not obliged to deposit, and will do so if, and only if, it is in their own best inter22

est. With this is mind, define a bank run as follows Definition 4.1 A bank run is a strategic situation where the unique Nash equilibrium in pure strategies of at least one of the subgames defining the post-deposit coordination environment faced by patient depositors is given by either (w, nw) or (w, w). Alternatively, a bank run is a strategic situation where at least one agent has a rational incentive to deposit funds in period zero and misrepresent her type in period one for a subset of post-deposit Bayesian coordination games, (w ) . Mean preserving information spreads on the expected state of the economy provide us with a simple setting to study self-fulfilling, as well as self-contained, bank runs. The following section illustrates numerically the main findings of this paper.
4.1 NUMERICAL PROOFS

This section presents a series of simulations of the pre and post-deposit game-theoretical strategic structures illustrating the existence of rational self-contained, as well as general, run equilibria for a given set of expected states observed by the bank, b (jointly with f (b ) ), and different sets of subjective beliefs hold by depositors, f a (a ) . The framework considered is rich enough to include states of the world generating attack prone and safe areas, allowing for a simple comparison with the currency 23 crises literature. We consider a framework with 100 agents (potential depositors), a unique bank, an exogenously fixed (as of period zero) interest rate of 5 percent, i.e. R=1.05, and a coefficient of relative risk aversion given by =2. It should be noted that none of the main results obtained relies on the particular values assigned to these variables, which remain within the reasonable limits defined in the literature, see Peck and Shell (2003), and Chari and Jagannathan (1988). Unless stated otherwise, the true set of expected states observed by the bank is given by b ={30, 50}, with its corresponding associated set of probabilities f (b ) ={0.8, 0.2}. These state values are chosen without loss of generality, and any mean preserving spread over the set of expected states defined in a
23

We are not the first ones to remark the importance of information differentials regarding current issues in international economics. Information heterogeneity among agents has been assumed by Morris and Shim (1998) to explain currency crises, Calvo and Mendoza (2000) to analyze the portfolio diversification and financial contagion phenomena, and van Wincoop and Bacchetta (2003) to study the exchange rate determination puzzle.

Simulation details and Matlab codes are available from the author upon request.

22

setting with multiple (>2) sequential allocations gives place to similar attack areas and does not affect the results obtained. In other words, runs do not depend on the number of contracts, but on the information spread suffered by depositors. This is not to say that we can always justify runs based on beliefs differentials independently of the number of allocations. For instance, not all demand deposit contracts allow for the existence of runs, either general or self-contained. Besides, even though attacks can be generated in a framework with N allocations, one per agent, it is unrealistic to assume that are triggered by information differentials between the bank and its depositors. Sunspots provide a much better justification in this case. The optimal sequential allocations composing the mechanism for b ={30, 50}, with f (b ) ={0.8, 0.2}, are given by
1 b c1 (1 ) =1.0162
b with associated utility u (c11 (1 )) =-

The subjective payoff structure defined by potential depositors relies heavily on the chosen probability function, f a (a ) . While this idea is intuitively clear, the depositing strategic framework conditional on expected selfcontained runs is entirely based on the relative realizations and probabilities composing the set of expected states observed by depositors. That is, the existence of self-contained run areas and the respective depositing decisions of agents depend on the relative expected state of the economy. Proposition 4.2 If a mechanism is vulnerable to general runs, there exists an interval of withdrawing depositors, which we refer to as self-contained run area, such that, any mean preserving spread probability function on the state of the economy, whose support includes a subset of expected realizations defined within this interval, may lead to a selfcontained run on the bank. Alternatively, any set of beliefs that, after observing w withdrawing agents, leads to an expected utility falling within the values that correspond to the variables delimiting the self-contain run area, may trigger a self-contained run on the bank. Proof. This proposition follows directly from Theorem 3.5. We calculate here the exact values defining the self-contained run area for b m* ( f (b )) , with ={30, 50} and f (b ) ={0.8, 0.2}. Consider again the utility derived from the optimal consumption allocations composing the mechanism
1 b u (c1 (1 )) =-0.984058,

0.984058, c12 (b ) =1.0105 with associated utility u (c12 (b )) =2 2 0.989609. The expected utility derived from remaining in autarky, if agents use the same information set as the bank to calculate their expected payoffs, equals

eu(aut)=-0.968571,
which is lower than the expected utility obtained from depositing

eu(dp)=-0.968446.
If no run is expected in period one, or is expected with low enough probability, agents have an incentive to deposit their funds in the bank, given identical information sets. Assume that agents observe a mean preserving spread state set defined over the true expected state set of the economy. That is, given b and a f (b ) , agents observe and f a (a ) , such that
b a ,
f ( b ) f a ( a ) ,

u (c12 (b )) =-0.989609. 2

The lower and upper bounds delimiting the interval that gives place to the self-contained run area, [m , M ] , are based on the above utility values
1 b b [100 30 c1 ( 1 ) ( m 1 ) c12 ( b )] R 2 m 70 m : 1 1

1 b = u ( c1 ( 1 ))

and

i =1,...,#b

f (b )b = i i

j =1,...,#a

f a (aj )aj ,

a j

1 b b [100 30c1 (1 ) (M 1 )c12 (b )]R 2 M 70 M : 1

= u (c12 (b )) 2

with b b , i = 1,..., # b , and i

, j = 1,..., # .
a a

The solutions to both equations are given by m =71.3988 and M =74.6254. Note that for an attack to be self-contained the maximum

23

number of patient agents expected to misrepresent their type cannot exceed the width of 24 the self-contained run interval. Any subjective set of beliefs whose support includes a subset of realizations defined within the interval [72,74], or giving place (after observing w withdrawing agents) to an expected utility (from reporting truthfully) that falls between 1 b u (c1 (1 )) and u (c12 (b )) , may trigger a self2 contained run on the bank. Theorem 4.3 If a mechanism is vulnerable to general runs, there exists a set of mean preserving spread probability functions defined on the expected state of the world, b and f (b ) , such that the subjective realization sets and beliefs these spreads give place to, a and f a (a ) , lead to self-contained attacks on the bank. Given the expected self-contained run that could take place in the interim period, agents deposit their funds in period zero. That is, the expected utility derived from depositing conditional on the possible self-contained run is strictly higher than the expected utility from remaining in autarky. Proof. It must be emphasized that, while the self-contained run area is defined by the interval [m , M ] presented in Proposition 4.2, selfcontained runs do not take place within this interval, whose effect is restricted to the updated Bayesian beliefs of depositors, but in the attack area defined by (ia , b ] (the equality in i subindexes is based on the numerical examples illustrated through the proof). We assume that patient depositors expecting a selfcontained run are not able to withdraw as soon as ia is realized, but must wait to observe ia + 1 agents in line before attacking. That is, patient agents cannot anticipate the realization of the state before receiving the corresponding signal through w . Consider the self-contained run area described in Proposition 4.2. As already stated, for the run to be self-contained, the maximum number of depositors expected to attack cannot exceed the width of [72, 74]. With this restriction in mind, define the following mean preserving spread on the true expected state
a ={28, 72} with f (a ) ={0.8636, 0.1364}.

Given this information set, the rhs of the cec calculated for the first 28 withdrawing agents delivers an expected utility value of
a nr eu(cec(1 )) = E[u (c2 (w 28 | f a (a )))] = a 1 b [ N 1 c1 (1 )]R a + f (1 | p, w 28)u a N 1 b 1 b b [ N 1 c1 (1 ) ((a + 1) 1 )c12 (b )]R 2 2 f (a | p, w 28)u 2 a N (2 + 1)

= 0.960053
b which is higher than u (c11 (1 )) =-0.984058, so no attack takes place. The second term defined in a eu (cec(1 )) must account for the effect that a self-contained run would have on the final number of expected withdrawing depositors and the corresponding consumption allocation. However, f (a | p, w 28) must be calcu2 lated for a =72 to be consistent with the sub2 jective beliefs of depositors.

If an additional agent is observed withdrawing after 28 depositors have withdrawn funds, the subjective beliefs of the remaining patient depositors must be updated accordingly, which modifies the expected utility value obtained from the rhs of the cec to
nr eu (cec(a )) = E[u (c2 (w > 28 | f a ( a )))] =-0.986639 2

u (c12 (b )) =-0.989609 2

b < eu(cec(a )) < u (c11 (1 )) = 2

-0.984058, triggering a self-contained run on the bank, see the Self-Contained Attack Annex 1. Given the self-contained run that would take place if a is realized, agents deposit in period zero iff 2 the expected utility derived from depositing
a 1 1 b N u ( c 1 ( 1 )) + a + eu ( dp ) = f a ( 1 ) N a 1 a 1 u ( c 2 ( 1 )) N b b 1 ( a + 1) 1 1 b 2 N u ( c 1 ( 1 )) + N fa ( a ) 2 N ( a + 1) 2 2 u ( c 2 ( a + 1 )) 2 N

u ( c 12 ( b )) + 2

is strictly higher than the expected utility from remaining in autarky

24

Though necessary, this condition is clearly not sufficient.

24

a a N 1 a eu(aut) = f a (1 ) 1 u(1) + N N u(R) +

nous) stochastic variable defined within the homogeneous information spread framework studied, it can be easily shown that Proposition 5.1 In an environment with either one or multiple banks, suspension of convertibility prevents all self-contained runs based on information differentials if it is announced after agents deposit. Moreover, suspension eliminates any run based on the realization of an exogenous sunspot variable if # 1 b c2 (b ) c1 (1 ) . #
b b

a N a 2 u(R) f a (a ) 2 u(1) + 2 N N
This condition is satisfied for a ={28, 72} and f (a ) ={0.8636, 0.1364}, since eu(dp)=-0.968537 > eu(aut)=-0.968572. We perform the same analysis for an alternative expected state of the world, given by b ={20, 40} and f (b ) ={0.8, 0.2}. The optimal consumption allocations composing the mechanism are
1 b c1 (1 ) =1.0187 b with associated utility u (c11 (1 )) =

If suspension is not announced and a sunspotbased general run is expected to take place, the maximum probability of such an event that allows for agents to deposit is given by the solution to

-0.981643, c12 (b ) =1.0137 with associated utility u (c12 (b )) = 2 2 -0.986485, defining a self-contained run interval delimited by m =66.2142 and M =69.5472. Consider the following mean preserving spreads on the expected state of the economy, and the corresponding depositing decisions of agents
a ={18, 67} with f (a ) ={0.8776, 0.1224} eu(dp)=-0.963791 > eu(aut)=-0.963808 a ={18, 68} with f (a ) ={0.88, 0.12} eu(dp)=-0.963793 > eu(aut)=-0.963810

f(nr)E[U(n r)]+f(run) E[U(run)] E[U(aut)] , (3)

where f(run) defines the subjective probability assigned to the realization of an exogenous run-triggering sunspot variable, and f(nr)=1f(run). Given N=100, b ={30, 50} and f (b ) ={0.8, 0.2}, this probability is equal to 0.002641 if both last agents in line are patient, 25 and 0.0025145 if they are both impatient. Any agent expecting a sunspot-based run to take place with a subjective probability higher than 0.002641 will not deposit funds, unless suspension is imposed and constitutes an efficient containment policy. Proposition 5.2 In an environment with ei-

Agents deposit despite the self-contained run that would take place if a is realized. 2

5. A note on Suspension of Convertibility

If suspension is announced as part of m* ( f (b )) in the above numerical examples, all agents expecting realizations of the random variable higher than b will not deposit funds, since a 2 payment of zero is received with positive probability. However, the mean preserving spread distribution defined on the true set of expected states may not be wide enough to trigger a self-contained run. In this case, agents would deposit if the bank does not announce suspension, but do not expect or generate a (self-contained) run in the interim period. Therefore, given the unique (endoge-

ther one or multiple banks, suspension of convertibility prevents all sunspot-based runs if it is announced before agents deposit and
# 1 b c2 (b b ) c1 (1 ) #
b

On the other hand, any bank imposing an efficient suspension policy as part of m* ( f (b )) attracts all agents whose f(run) is higher than 0.002641, as equation (3) would be satisfied for any subjective value assigned to f(run). Thus, given a unique endogenous stochastic variable, multiple banks, and an efficient suspension policy, banks should coordinate the timing of their announcements in order to implement the optimal mechanism among all
25

An equal distribution policy between the depositors remaining

after

is reached has been assumed.

25

agents. The current theoretical framework supports such an optimality requirement due to its simplicity and the fact that it consists of a unique banking cycle. Assume that a second (endogenous) stochastic variable, given by R, is added to the current model such that the allocations composing the mechanism are also based on its expected value. The realization of R is assumed to be observed in the final (second) period, but interim signals about its expected value are received by a subset of depositors, as proposed

by Jacklin and Bhattacharya (1988). Clearly, our theoretical framework is compatible with this extension, within which suspension does not constitute an efficient containment policy per se, since it could be used to contain runs generated by a fundamental variable. At the same time, modifying the assumed utility function to allow for the existence of depositors with highly biased sunspot probabilities would provide us with a formal strategic scenario where the containment of fundamental, sunspot and information based runs determines the dynamic stability of the banking system.

26

Annex 1. Self-Contained Attack

27

Bibliographical references
Adao, B. and T. Temzelides, Sequential Equilibrium and Competition in a Diamond-Dybvig Banking Model, Review of Economic Dynamics, 1, (1998), 859-877. Alonso, Irasema, Truthful Revelation in the Diamond and Dybvig Banking Environment, Journal of Mone-

tary Economics, 37, (1996), 73-87.

Calvo, G. A. and E. G. Mendoza, Rational Contagion and the Globalization of Securities Markets, Journal of International Economics, 51, (2000), 79-113. Chari, V.V. and R. Jagannathan. Banking Panics, Information, and Rational Expectations Equilibrium,

Journal of Finance, 43, (1988), 749-761.

Cole, H.L. and T. Kehoe. Self-Fulfilling Debt Crises, Review of Economic Studies, 67 (1), (2000), 91-116. Diamond, D. and P. Dybvig. Bank Runs, Deposit Insurance, and Liquidity, Journal of Political Economy, 91, (1983), 401-419. Furman, J. and J.E. Stiglitz. Economic Crises: Evidence and Insights from East Asia, Brookings Papers on Economic Activity, 2, (1998), 1-135. Green, E.J. and P. Lin. Diamond and Dybvig's Classic Theory of Financial Intermediation: What's Missing? Federal Reserve Bank of Minneapolis Quarterly Review, 24, (Winter 2000), 3-13. Green, E.J. and P. Lin. Implementing Efficient Allocations in a Model of Financial Intermediation, Journal of Economic Theory, 109, (2003), 1-23. Jacklin, Charles. Demand Deposits, Trading Restrictions, and Risk Sharing, In: Contractual Arrangements for Intertemporal Trade, ed. Edward C. Prescott and Neil Wallace, pp. 26-47. Minnesota Studies in Macroeconomics, Vol. 1. University of Minnesota Press, (1987). Jacklin, C. and S. Bhattacharya. Distinguishing Panics and Information-based Bank Runs: Welfare and Policy Implications, Journal of Political Economy, 96, (1988), 568-592. Kaminsky, G.L. and C.M. Reinhart. The Twin Crises: The Causes of Banking and Balance-Of-Payments Problems, American Economic Review, 89, (June 1999), 473-500. Kahneman, D. and A. Tversky. Prospect Theory - Analysis of Decision under Risk, Econometrica, 47 (2), (1979), 263-291. Krasa, S. and A.P. Villamil. Monitoring the Monitor. An Incentive Structure for a Financial Intermediary, Journal of Economic Theory, 57, (June 1992), 197-221. Krugman, Paul. A Model of Balance-of-Payments Crises, Journal of Money, Credit, and Banking, 11, (1979), 311-325. Morris, S. and H.S. Shin. Unique Equilibrium in a Model of Self-Fulfilling Currency Attacks, American Economic Review, 88, (June 1998), 587-597. Obstfeld, Maurice. The Logic of Currency Crises, Cahiers Economiques et Monetaires (Banque de France), 43, (1994), 189-213. Peck, J. and K. Shell. Equilibrium Bank Runs, Journal of Political Economy, 111, (2003), 103-123. Postlewaite, A. and X. Vives. Bank Runs as an Equilibrium Phenomenon, Journal of Political Economy, 95, (1987), 485-491. Samartin, Margarita. Should Bank Runs be Prevented?, Journal of Banking and Finance, 27, (2003), 9771000. Tversky, A. and D. Kahneman. Advances in Prospect Theory - Cumulative Representation of Uncertainty, Journal of Risk and Uncertainty, 5 (4), (1992), 297-323.

28

van Wincoop, E. and P. Bacchetta. Can Information Heterogeneity Explain the Exchange Rate Determination Puzzle?, NBER Working Paper No. 9498, (February 2003)

29

ltimos ttulos publicados

DOCUMENTOS DE TRABAJO EL VALOR ECONMICO DEL ESPAOL
DT 14/10 DT 13/08 DT 12/08 DT 11/08 DT 10/08 DT 09/07 DT 08/07 DT 07/07 DT 06/07 DT 05/07 DT 04/07 DT 03/06 DT 02/06 DT 01/06 Antonio Alonso, Jos; Gutirrez, Rodolfo: Lengua y emigracin: Espaa y el espaol en las

migraciones internacionales.
de Diego lvarez, Dorotea; Rodrigues-Silveira, Rodrigo; Carrera Troyano Miguel: Estrate-

gias para el Desarrollo del Cluster de Enseanza de Espaol en Salamanca

Quirs Romero, Cipriano: Lengua e internacionalizacin: El papel de la lengua en la internacionalizacin de las operadoras de telecomunicaciones. Girn, Francisco Javier; Caada, Agustn: La contribucin de la lengua espaola al PIB y al

empleo: una aproximacin macroeconmica.

Jimnez, Juan Carlos; Narbona, Aranzazu: El espaol en el comercio internacional. Carrera, Miguel; Ogonowski, Micha: El valor econmico del espaol: Espaa ante el espejo

de Polonia.
Rojo, Guillermo: El espaol en la red. Carrera, Miguel; Bonete, Rafael; Muoz de Bustillo, Rafael: El programa ERASMUS en el

marco del valor econmico de la Enseanza del Espaol como Lengua Extranjera.
Criado, Mara Jess: Inmigracin y poblacin latina en los Estados Unidos: un perfil socio-

demogrfico.
Gutirrez, Rodolfo: Lengua, migraciones y mercado de trabajo. Quirs Romero, Cipriano; Crespo Galn, Jorge: Sociedad de la Informacin y presencia del

espaol en Internet.
Moreno Fernndez, Francisco; Otero Roth, Jaime: Demografa de la lengua espaola. Alonso, Jos Antonio: Naturaleza econmica de la lengua. Jimnez, Juan Carlos: La Economa de la lengua: una visin de conjunto.

WORKING PAPERS WP 02/10 WP 01/10 WP 14/09 WP 13/09 WP 12/09 WP 11/09 WP 10/09 Santos-Arteaga, Francisco J.: Bank Runs Without Sunspots. Donoso, Vicente; Martn, Vctor: La sostenibilidad del dficit exterior de Espaa. Dobado, Rafael; Garca, Hctor: Neither so low nor so short! Wages and heights in eighteenth and early nineteenth centuries colonial Hispanic America. Alonso, Jos Antonio: Colonisation, formal and informal institutions, and development. lvarez, Francisco: Oportunity cost of CO2 emission reductions: developing vs. developed economies. J. Andr, Francisco: Los Biocombustibles. El Estado de la cuestin. Luengo, Fernando: Las deslocalizaciones internacionales. Una visin desde la economa

crtica

30

WP 09/09 WP 08/09 WP 07/09 WP 06/09 WP 05/09 WP 04/09 WP 03/09 WP 02/09 WP 01/09 WP 08/08

Dobado, Rafael; Guerrero, David: The Integration of Western Hemisphere Grain Markets in

the Eighteenth Century: Early Progress and Decline of Globalization.

lvarez, Isabel; Marn, Raquel; Maldonado, Georgina: Internal and external factors of com-

petitiveness in the middle-income countries.

Minondo, Asier: Especializacin productiva y crecimiento en los pases de renta media. Martn, Vctor; Donoso, Vicente: Seleccin de mercados prioritarios para los Pases de Renta

Media.
Donoso, Vicente; Martn, Vctor: Exportaciones y crecimiento econmico: estudios empri-

cos.
Minondo, Asier; Requena, Francisco: Qu explica las diferencias en el crecimiento de las exportaciones entre los pases de renta media? Alonso, Jos Antonio; Garcimartn, Carlos: The Determinants of Institutional Quality. More

on the Debate.
Granda, Ins; Fonfra, Antonio: Technology and economic inequality effects on interna-

tional trade.
Molero, Jos; Portela, Javier y lvarez Isabel: Innovative MNEs Subsidiaries in different

domestic environments.
Boege, Volker; Brown, Anne; Clements, Kevin y Nolan Anna: Qu es lo fallido? Los

Estados del Sur,o la investigacin y las polticas de Occidente? Un estudio sobre rdenes polticos hbridos y los Estados emergentes.
WP 07/08 WP 06/08 WP 05/08 WP 04/08 WP 03/08 WP 02/08 WP 01/08 WP 05/07 WP 04/07 WP 03/07 WP 02/07 WP 01/07 WP 05/06 Medialdea Garca, Bibiana; lvarez Peralta, Nacho: Liberalizacin financiera internacional,

inversores institucionales y gobierno corporativo de la empresa

lvarez, Isabel; Marn, Raquel: FDI and world heterogeneities: The role of absorptive ca-

pacities
Molero, Jos; Garca, Antonio: Factors affecting innovation revisited Tezanos Vzquez, Sergio: The Spanish pattern of aid giving Fernndez, Esther; Prez, Rafaela; Ruiz, Jess: Double Dividend in an Endogenous Growth

Model with Pollution and Abatement

lvarez, Francisco; Camia, Ester: Moral hazard and tradeable pollution emission permits. Cerd Tena, Emilio; Quiroga Gmez, Sonia: Cost-loss decision models with risk aversion. Palazuelos, Enrique; Garca, Clara: La transicin energtica en China. Palazuelos, Enrique: Dinmica macroeconmica de Estados Unidos: Transicin entre dos

recesiones?
Angulo, Gloria: Opinin pblica, participacin ciudadana y poltica de cooperacin en

Espaa.
Luengo, Fernando; lvarez, Ignacio: Integracin comercial y dinmica econmica: Espaa

ante el reto de la ampliacin.

lvarez, Isabel; Magaa, Gerardo: ICT and Cross-Country Comparisons: A proposal of a

new composite index.

Schnemann, Julia: Cooperacin interregional e interregionalismo: una aproximacin so-

cial-constructivista.

31

WP 04/06 WP 03/06 WP 02/06

Kruijt, Dirk: Amrica Latina. Democracia, pobreza y violencia: Viejos y nuevos actores. Donoso, Vicente; Martn, Vctor: Exportaciones y crecimiento en Espaa (1980-2004): Cointegracin y simulacin de Montecarlo. Garca Snchez, Antonio; Molero, Jos: Innovacin en servicios en la UE: Una aproximacin

a la densidad de innovacin y la importancia econmica de los innovadores a partir de los datos agregados de la CIS3.
WP 01/06 WP 06/05 WP 05/05 WP 04/05 WP 03/05 WP 02/05 WP 01/05 Briscoe, Ivan: Debt crises, political change and the state in the developing world. Palazuelos, Enrique: Fases del crecimiento econmico de los pases de la Unin Europea

15.
Leyra, Begoa: Trabajo infantil femenino: Las nias en las calles de la Ciudad de Mxico. lvarez, Isabel; Fonfra, Antonio; Marn Raquel: The role of networking in the competitive-

ness profile of Spanish firms.

Kausch, Kristina; Barreada, Isaas: Alliance of Civilizations. International Security and

Cosmopolitan Democracy.
Sastre, Luis: An alternative model for the trade balance of countries with open economies:

the Spanish case.

Daz de la Guardia, Carlos; Molero, Jos; Valadez, Patricia: International competitiveness in

services in some European countries: Basic facts and a preliminary attempt of interpretation.
WP 03/04 WP 02/04 WP 01/04 Angulo, Gloria: La opinin pblica espaola y la ayuda al desarrollo. Freres, Christian; Mold, Andrew: European Union trade policy and the poor. Towards im-

proving the poverty impact of the GSP in Latin America.

lvarez, Isabel; Molero, Jos: Technology and the generation of international knowledge spillovers. An application to Spanish manufacturing firms.

POLICY PAPERS PP 03/09 PP 02/09 PP 01/09 PP 04/08 Ferguson, Lucy: Analysing the Gender Dimensions of Tourism as a Development Strategy. Carrasco Gallego ,Jos Antonio: La Ronda de Doha y los pases de renta media. Rodrguez Blanco, Eugenia: Gnero, Cultura y Desarrollo: Lmites y oportunidades para el cambio cultural pro-igualdad de gnero en Mozambique. Tezanos, Sergio: Polticas pblicas de apoyo a la investigacin para el desarrollo. Los casos

de Canad, Holanda y Reino Unido

PP 03/08 PP 02/08 PP 01/08 Mattioli, Natalia Including Disability into Development Cooperation. Analysis of Initiatives

by National and International Donors

Elizondo, Luis: Espacio para Respirar: El humanitarismo en Afganistn (2001-2008). Carams Boada, Albert: Desarme como vnculo entre seguridad y desarrollo. La reintegra-

cin comunitaria en los programas de Desarme, desmovilizacin y reintegracin (DDR) de combatientes en Hait.
PP 03/07 PP 02/07 PP 01/07 Guimn, Jos: Government strategies to attract R&D-intensive FDI. Czapliska, Agata: Building public support for development cooperation. Martnez, Ignacio: La cooperacin de las ONGD espaolas en Per: hacia una accin ms

estratgica.

32

PP 02/06 PP 01/06 PP 04/05 PP 03/05 PP 02/05 PP 01/05 PP 02/04 PP 01/04

Ruiz Sandoval, Erika: Latinoamericanos con destino a Europa: Migracin, remesas y codesa-

rrollo como temas emergentes en la relacin UE-AL.

Freres, Christian; Sanahuja, Jos Antonio: Hacia una nueva estrategia en las relaciones

Unin Europea Amrica Latina.

Manalo, Rosario; Reyes, Melanie: The MDGs: Boon or bane for gender equality and wo-

mens rights?
Fernndez, Rafael: Irlanda y Finlandia: dos modelos de especializacin en tecnologas avan-

zadas.
Alonso, Jos Antonio; Garcimartn, Carlos: Apertura comercial y estrategia de desarrollo. Lorente, Maite: Dilogos entre culturas: una reflexin sobre feminismo, gnero, desarrollo y

mujeres indgenas kichwuas.

lvarez, Isabel: La poltica europea de I+D: Situacin actual y perspectivas. Alonso, Jos Antonio; Lozano, Liliana; Prial, Mara ngela: La cooperacin cultural espa-

ola: Ms all de la promocin exterior.

33